Unverified50% confidenceFactExact time
GCD uses the model's own Jacobian matrix as a constraint to solve an optimization problem, making the direction of the logit gradient exposed to the attacker maximally misaligned with the direction of the true adversarial loss gradient.
1
Sources
50%
Confidence
Long-term
Relevance
10/14/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedThe augmented Lagrangian method explicitly models the inter-layer relationships in predictive coding as a constrained optimization problem, retaining penalty terms while introducing Lagrangian multipliers to satisfy the constraints.69% similarUnverified白盒攻击（如FGSM和PGD）假设攻击者能够访问模型的完整参数，通过反向传播计算最优扰动方向66% similarUnverifiedGFlowNets model the generation process as a flow problem on a directed acyclic graph (DAG), training for flow conservation so that different terminal states are sampled with probability proportional to their reward.63% similarUnverifiedThe design of decoding strategies can make more precise use of the log-linear geometric structure of logits rather than relying on rough distributional assumptions.63% similarUnverifiedReasoning chain outputs have become the most valuable collection target for adversarial distillation, because attackers do not need to reproduce the expensive RL training process and can instead use the collected reasoning trajectories as supervised signals to distill models with strong reasoning capabilities.63% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1023777API
curl https://kongchang.com/api/v1/knowledge/claims/1023777MCP
get_claim(id=1023777)